nowadays, when you see a tutorial with Indian accent, you know it has better chances to be a good one. This time it didn't disappoint. Kudos to Indian engineering, Kudos to Dr. Verma
والله كلامك جميل جدا وسهل لقد فهمت كل شي عن ANFIS MODEL. I'm Shahed, a master's student in Jordan; thank u so much. im speak Arabic but i wrote in English to express to u
I have a question. Why does the form of fi in layer 4 give the formula of px+qy+r? If the input value is added (e.g. z) Does the formula of px+qy+kz+r? (k is parameter)
Dr. Verma, I am working on a project to control a small wind turbine's blade pitch angle to achieve maximum wind energy efficiency without exceeding 47 volts of output. The main inputs for the control system are voltage and RPM. Additionally, the turbine will be enclosed in a box, and the wind will be manually controlled by a person. The problem is that I don't have a wind speed sensor because they are too big and I have to infer the wind speed with the RPM. Given these conditions, what type of control system would you recommend to optimize the blade pitch angle? Would a PID controller, a Fuzzy Logic Controller, a ANFIS ,or another approach be more suitable?
Fuzzy system works on pre defined IF-THEN rules. Once they are defined, they are not changed. While Neural network has weights which are updated with training.
yes, there should have been 4 nodes. We should have one rule for every combination of formulas, as we can see on the picture at 17:45. So, the missed nodes would contain A1 X B2 and A2 X B1
I'm curious about the reason why we don't have to write rules for fuzzy inference. Does ANFIS work with no rules? Or, does it create rules automatically? (If ANFIS work with rules, does 'fuzzy logic' have its own significance? )
Athough I enjoyed this presentation as I got to know details about the learning process that I didn't know before, my reaction was similar to yours. The primary purpose of using the ANFIS is probably not to use it as a universal function approximator (standard ANNs are probably a better choice for that, both in terms of flexibility and efficiency), but to explicitly exploit domain knowledge through a fuzzy rule set. What we should be able to hope for is 1) to control the learning to conform to domain knowledge and 2) to exploit training data to fine tune the domain knowledge. And I still believe (without having tried it out, admittedly) that can be achieved. I searched the Internet and it seems that there is an addRule function that can be used for this purpose. From this presentation, interestingly, it seems that one does not have to specify rules, in which case you just get some free parameters (embedded in fuzzy rule structures) which can be learned during the training (in a very specific ANN structure). But if you extract the rules in the end (after training), it seems unlikely that they will contain much semantic information.
nowadays, when you see a tutorial with Indian accent, you know it has better chances to be a good one. This time it didn't disappoint. Kudos to Indian engineering, Kudos to Dr. Verma
Found your video at the perfect time when I needed the most. Thank you
والله كلامك جميل جدا وسهل لقد فهمت كل شي عن ANFIS MODEL.
I'm Shahed, a master's student in Jordan; thank u so much. im speak Arabic but i wrote in English to express to u
Nicely Done!.... Clear and detailed. Please keep making this type of video.
excellent video, thank you
Well explained...
Thank you so much Dr. This 30++ minutes video has given me 50% understanding on ANFIS theory & implementation.
Surely a useful presentation.
Useful content.
Brilliant🎉
Really valuable 👍
Super presentation
Thank you so much, sir. I've discovered your videos right when I needed to build fuzzy logic applications in Matlab.
thank you ❤ Dr
Thank your sir
Thanks a lot. well explained very useful
I have a question. Why does the form of fi in layer 4 give the formula of px+qy+r?
If the input value is added (e.g. z) Does the formula of px+qy+kz+r? (k is parameter)
Sir kindly make a video on 'Implemetation of ANFIS For fault detection in power system.
very good
Thank you for the great lecture. Please I need detail training. And application to control system.
THANKYOU SIR
Nice explanation, thank you very much
How do we know whether the model was overfitting or not?
We can also check for overfitting. But that, I have not explained.
Sir is there any advance version of ANFIS which have less computational complexity and we can apply on sensor data to get predictions
Sir matlab ki full video upload kigiye
Dr. Verma, I am working on a project to control a small wind turbine's blade pitch angle to achieve maximum wind energy efficiency without exceeding 47 volts of output. The main inputs for the control system are voltage and RPM. Additionally, the turbine will be enclosed in a box, and the wind will be manually controlled by a person. The problem is that I don't have a wind speed sensor because they are too big and I have to infer the wind speed with the RPM.
Given these conditions, what type of control system would you recommend to optimize the blade pitch angle? Would a PID controller, a Fuzzy Logic Controller, a ANFIS ,or another approach be more suitable?
Simplest is PID. It can do well. But ANFIS can be tried. Compare both. It will give u one research paper.
thanks u, very easy to understand, althought my eng is not good
thank for your video
I have a question, from which paper you said that FIS are not able to learn?
I really need the refrence
Fuzzy system works on pre defined IF-THEN rules. Once they are defined, they are not changed.
While Neural network has weights which are updated with training.
@@exploringtechnologies9 thank for your response, I understand it.
Can u share how design Reactive current PI-fuzzy for a three-phase grid-connected PV
Can hou clarify why we only need 2 nodes in layer 2? Shouldn't it be 4 since its 2^2? Thanks!
did you get why it is like that? i have the same doubt
@@onkarchaudhary1233 me too bro. I think the equation RF = (mf)^n is not corrected. It should be RF = mf * n
yes, there should have been 4 nodes. We should have one rule for every combination of formulas, as we can see on the picture at 17:45. So, the missed nodes would contain A1 X B2 and A2 X B1
I'm curious about the reason why we don't have to write rules for fuzzy inference. Does ANFIS work with no rules? Or, does it create rules automatically? (If ANFIS work with rules, does 'fuzzy logic' have its own significance? )
Athough I enjoyed this presentation as I got to know details about the learning process that I didn't know before, my reaction was similar to yours. The primary purpose of using the ANFIS is probably not to use it as a universal function approximator (standard ANNs are probably a better choice for that, both in terms of flexibility and efficiency), but to explicitly exploit domain knowledge through a fuzzy rule set.
What we should be able to hope for is 1) to control the learning to conform to domain knowledge and 2) to exploit training data to fine tune the domain knowledge. And I still believe (without having tried it out, admittedly) that can be achieved. I searched the Internet and it seems that there is an addRule function that can be used for this purpose.
From this presentation, interestingly, it seems that one does not have to specify rules, in which case you just get some free parameters (embedded in fuzzy rule structures) which can be learned during the training (in a very specific ANN structure). But if you extract the rules in the end (after training), it seems unlikely that they will contain much semantic information.
Can u make some data prediction ANFIS? With matlab?
Times series prediction video is already on my channel.
Useful content.